libavfilter/af_amix.c: mixing without volume renormalization

[CoRoNe]

It seems the amix filter always does "volume renormalization" and there's no way to turn that off.
I'd like to mix audio-segments at full volume and was actually looking for a similar approach as Avisynth's MixAudio(clip1,clip2,1.0,1.0) filter. No volume renormalization, just mixing.
According to Paul Mahol this isn't possible at the moment.
I'd like to request for this (imo obvious) feature to be implemented.

For reference see: https://ffmpeg.org/pipermail/ffmpeg-user/2019-July/044807.html.

[Cigaes]

Use amerge and pan.

[CoRoNe]

Please have a look at the reference url. This has been discussed already. I can't use these filters.

[Carl Eugen Hoyos]

All information should be here on the bug tracker, do not use external resources except for large sample files.

[CoRoNe]

Avisynth-script

[CoRoNe]

I have found no way to edit my first post, so...

About 7 years ago I've used Audacity to assemble the soundtrack of the videogame No One Lives Forever 2. See ​https://www.youtube.com/watch?v=4Y3aKcQ0HK4 for example.
The soundtrack comprised of lots of small segments that can all dynamically be loaded by the videogame.

A year later I've assembled the soundtrack once more, but this time with Avisynth. See the attached SIBERIA.avs for example.

Now I wanted to see if ffmpeg is up for the task. Just for fun.

FFmpeg used:

ffmpeg version N-94137-g89b9690-Reino Copyright (c) 2000-2019 the FFmpeg developers
  built with gcc 8.3.0 (GCC)
  configuration: --arch=x86 --target-os=mingw32 --cross-prefix=/cygdrive/m/ffmpeg-windows-build-helpers-master/ffmpeg_local_builds/sandbox/cross_compilers/mingw-w64-i686/bin/i686-w64-mingw32- --pkg-config=pkg-config --pkg-config-flags=--static --extra-version=Reino --enable-gray --enable-version3 --disable-debug --disable-doc --disable-htmlpages --disable-manpages --disable-podpages --disable-txtpages --disable-w32threads --enable-avisynth --enable-frei0r --enable-filter=frei0r --enable-gmp --enable-gpl --enable-libaom --enable-libass --enable-libbluray --enable-libbs2b --enable-libcaca --extra-cflags=-DCACA_STATIC --enable-libfdk-aac --enable-libflite --enable-libfontconfig --enable-libfreetype --enable-libfribidi --enable-libgme --enable-libgsm --enable-libilbc --enable-libmp3lame --enable-libmysofa --enable-libopencore-amrnb --enable-libopencore-amrwb --enable-libopenh264 --enable-libopenmpt --enable-libopus --enable-librubberband --enable-libsnappy --enable-libsoxr
--enable-libspeex --enable-libtheora --enable-libtwolame --extra-cflags=-DLIBTWOLAME_STATIC --enable-libvidstab --enable-libvmaf --enable-libvo-amrwbenc --enable-libvorbis --enable-libvpx --enable-libwebp --enable-libx264 --enable-libx265 --enable-libxavs --enable-libxml2 --enable-libxvid --enable-libzimg --enable-libzvbi --enable-mbedtls --extra-cflags='-march=pentium3' --extra-cflags='-mtune=athlon-xp' --extra-cflags=-O2 --extra-cflags='-mfpmath=sse' --extra-cflags=-msse --enable-static --disable-shared --prefix=/cygdrive/m/ffmpeg-windows-build-helpers-master/ffmpeg_local_builds/sandbox/cross_compilers/mingw-w64-i686/i686-w64-mingw32
  libavutil      56. 30.100 / 56. 30.100
  libavcodec     58. 53.100 / 58. 53.100
  libavformat    58. 28.101 / 58. 28.101
  libavdevice    58.  7.100 / 58.  7.100
  libavfilter     7. 55.100 /  7. 55.100
  libswscale      5.  4.101 /  5.  4.101
  libswresample   3.  4.100 /  3.  4.100
  libpostproc    55.  4.100 / 55.  4.100

The soundtrack has an "explore", "warning" and "combat" section.

"Explore":  'SIBERIAE{1-29}.WAV'
Transition: 'TRANSEWSTR100.WAV'
"Warning":  'SIBERIAW{1-22}.WAV'
Transition: 'TRANSWE19.WAV'
"Combat":   'SIBERIAC{1-25}.WAV'
Transition: 'TRANSCS21.WAV'

These segments are actually mp3 disguised as wav (blame the videogame).
Sadly it's not a matter of simply concatenating all segments to create the entire score, because each segment starts at a very specific moment in time and creates overlap with the rest of them. Then all of them need to be mixed (at full volume).
I'm actually looking for a way to do a MixAudio(clip1,clip2,1.0,1.0) ​(Avisynth) with FFmpeg.

First let's have a look at amix's options:

ffmpeg -h filter=amix
[...]
amix AVOptions:
  inputs            <int>        ..F.A.... Number of inputs. (from 1 to 1024) (default 2)
  duration          <int>        ..F.A.... How to determine the end-of-stream. (from 0 to 2) (default longest)
     longest                      ..F.A.... Duration of longest input.
     shortest                     ..F.A.... Duration of shortest input.
     first                        ..F.A.... Duration of first input.
  dropout_transition <float>      ..F.A.... Transition time, in seconds, for volume renormalization when an input stream ends. (from 0 to INT_MAX) (default 2)
  weights           <string>     ..F.A.... Set weight for each input. (default "1 1")

I figure weights=1 1 does the same 1.0,1.0 in MixAudio(clip1,clip2,1.0,1.0), so that's good.
But then it appears amix will always do "volume renormalization". You can change the dropout_transition-option some what, but there's no way to turn if off.

This is the command I'm using / had to use so far:

ffmpeg \
-i SIBERIAE1.WAV -i SIBERIAE2.WAV -i SIBERIAE3.WAV -i SIBERIAE4.WAV -i SIBERIAE5.WAV \
-i SIBERIAE6.WAV -i SIBERIAE7.WAV -i SIBERIAE8.WAV -i SIBERIAE9.WAV -i SIBERIAE10.WAV \
-i SIBERIAE11.WAV -i SIBERIAE12.WAV -i SIBERIAE13.WAV -i SIBERIAE14.WAV -i SIBERIAE15.WAV \
-i SIBERIAE16.WAV -i SIBERIAE17.WAV -i SIBERIAE18.WAV -i SIBERIAE19.WAV -i SIBERIAE20.WAV \
-i SIBERIAE21.WAV -i SIBERIAE22.WAV -i SIBERIAE23.WAV -i SIBERIAE24.WAV -i SIBERIAE25.WAV \
-i SIBERIAE26.WAV -i SIBERIAE27.WAV -i SIBERIAE28.WAV -i SIBERIAE29.WAV \
-i TRANSEWSTR100.WAV \
-i SIBERIAW1.WAV -i SIBERIAW2.WAV -i SIBERIAW3.WAV -i SIBERIAW4.WAV -i SIBERIAW5.WAV \
-i SIBERIAW6.WAV -i SIBERIAW7.WAV -i SIBERIAW8.WAV -i SIBERIAW9.WAV -i SIBERIAW10.WAV \
-i SIBERIAW11.WAV -i SIBERIAW12.WAV -i SIBERIAW13.WAV -i SIBERIAW14.WAV -i SIBERIAW15.WAV \
-i SIBERIAW16.WAV -i SIBERIAW17.WAV -i SIBERIAW18.WAV -i SIBERIAW19.WAV -i SIBERIAW20.WAV \
-i SIBERIAW21.WAV -i SIBERIAW22.WAV \
-i TRANSWE19.WAV \
-i SIBERIAC1.WAV -i SIBERIAC2.WAV -i SIBERIAC3.WAV -i SIBERIAC4.WAV -i SIBERIAC5.WAV \
-i SIBERIAC6.WAV -i SIBERIAC7.WAV -i SIBERIAC8.WAV -i SIBERIAC9.WAV -i SIBERIAC10.WAV \
-i SIBERIAC11.WAV -i SIBERIAC12.WAV -i SIBERIAC13.WAV -i SIBERIAC14.WAV -i SIBERIAC15.WAV \
-i SIBERIAC16.WAV -i SIBERIAC17.WAV -i SIBERIAC18.WAV -i SIBERIAC19.WAV -i SIBERIAC20.WAV \
-i SIBERIAC21.WAV -i SIBERIAC22.WAV -i SIBERIAC23.WAV -i SIBERIAC24.WAV -i SIBERIAC25.WAV \
-i TRANSCS21.WAV \
-filter_complex "
  [1]adelay=158792S|158792S[E2];[2]adelay=291139S|291139S[E3];[3]adelay=423476S|423476S[E4];
  [4]adelay=555820S|555820S[E5];[5]adelay=714633S|714633S[E6];[6]adelay=873439S|873439S[E7];
  [7]adelay=1058736S|1058736S[E8];[8]adelay=1244019S|1244019S[E9];[9]adelay=1376339S|1376339S[E10];
  [10]adelay=1508682S|1508682S[E11];[11]adelay=1667496S|1667496S[E12];[12]adelay=1826306S|1826306S[E13];
  [13]adelay=1932181S|1932181S[E14];[14]adelay=2118076S|2118076S[E15];[15]adelay=2255399S|2255399S[E16];
  [16]adelay=2364348S|2364348S[E17];[17]adelay=2499655S|2499655S[E18];[18]adelay=2587883S|2587883S[E19];
  [19]adelay=2720226S|2720226S[E20];[20]adelay=2852568S|2852568S[E21];[21]adelay=2984913S|2984913S[E22];
  [22]adelay=3073138S|3073138S[E23];[23]adelay=3161367S|3161367S[E24];[24]adelay=3249594S|3249594S[E25];
  [25]adelay=3336742S|3336742S[E26];[26]adelay=3421112S|3421112S[E27];[27]adelay=3576621S|3576621S[E28];
  [28]adelay=3788369S|3788369S[E29];[29]adelay=3787479S|3787479S[EW100];[30]adelay=3841308S|3841308S[W1];
  [31]adelay=4089467S|4089467S[W2];[32]adelay=4337597S|4337597S[W3];[33]adelay=4585740S|4585740S[W4];
  [34]adelay=4751180S|4751180S[W5];[35]adelay=4916595S|4916595S[W6];[36]adelay=5040665S|5040665S[W7];
  [37]adelay=5206096S|5206096S[W8];[38]adelay=5309485S|5309485S[W9];[39]adelay=5412881S|5412881S[W10];
  [40]adelay=5516287S|5516287S[W11];[41]adelay=5619668S|5619668S[W12];[42]adelay=5785089S|5785089S[W13];
  [43]adelay=5950523S|5950523S[W14];[44]adelay=6033235S|6033235S[W15];[45]adelay=6198660S|6198660S[W16];
  [46]adelay=6364091S|6364091S[W17];[47]adelay=6446814S|6446814S[W18];[48]adelay=6598676S|6598676S[W19];
  [49]adelay=6722118S|6722118S[W20];[50]adelay=6872561S|6872561S[W21];[51]adelay=7038663S|7038663S[W22];
  [52]adelay=7204260S|7204260S[WE19];[53]adelay=7286129S|7286129S[C1];[54]adelay=7534279S|7534279S[C2];
  [55]adelay=7783363S|7783363S[C3];[56]adelay=8030562S|8030562S[C4];[57]adelay=8195988S|8195988S[C5];
  [58]adelay=8361416S|8361416S[C6];[59]adelay=8485486S|8485486S[C7];[60]adelay=8650917S|8650917S[C8];
  [61]adelay=8754307S|8754307S[C9];[62]adelay=8857700S|8857700S[C10];[63]adelay=8961094S|8961094S[C11];
  [64]adelay=9064486S|9064486S[C12];[65]adelay=9271272S|9271272S[C13];[66]adelay=9436697S|9436697S[C14];
  [67]adelay=9519413S|9519413S[C15];[68]adelay=9726192S|9726192S[C16];[69]adelay=9850266S|9850266S[C17];
  [70]adelay=9932983S|9932983S[C18];[71]adelay=10181124S|10181124S[C19];[72]adelay=10429267S|10429267S[C20];
  [73]adelay=10677408S|10677408S[C21];[74]adelay=10842834S|10842834S[C22];[75]adelay=10946226S|10946226S[C23];
  [76]adelay=11049615S|11049615S[C24];[77]adelay=11153012S|11153012S[C25];[78]adelay=11401044S|11401044S[CS21];
  [0][E2][E3][E4][E5][E6][E7][E8][E9][E10][E11][E12][E13][E14][E15][E16][E17][E18][E19][E20][E21][E22][E23]
  [E24][E25][E26][E27][E28][E29]
  [EW100]
  [W1][W2][W3][W4][W5][W6][W7][W8][W9][W10][W11][W12][W13][W14][W15][W16][W17][W18][W19][W20][W21][W22]
  [WE19]
  [C1][C2][C3][C4][C5][C6][C7][C8][C9][C10][C11][C12][C13][C14][C15][C16][C17][C18][C19][C20][C21][C22][C23]
  [C24][C25]
  [CS21]
  amix=inputs=79:dropout_transition=270.28,volume=37.3dB
" -f wav <some output>

• With the adelay filter all segments start at a very specific moment in time. (having to specify the delay for both channels is rather cumbersome, hence my ​one delay-string for all channels request)
• I do -filter_complex "[...]amix=inputs=79" -f null - to get the duration of the entire score: time=00:04:30.28 to enter as parameter for the dropout_transition-option. I want the volume to be left untouched, but since there's no way to turn "volume renormalization" off at the moment, entering the duration for dropout_transition appears to do the least damage.
• This obviously still results in a very low volume, so I do -filter_complex "[...]amix=inputs=79:dropout_transition=270.28,volumedetect" -f null -: mean_volume: -59.3 dB, max_volume: -37.3 dB...
• ...to crank up the volume to (I think) where it was before: -filter_complex "[...]amix=inputs=79:dropout_transition=270.28,volume=37.3dB".

Mixing audio-segments at full volume without any kind of normalization is a rather basic feature, or at least it should be in my opinion.
Therefore I request "volume renormalization" to be optional and that you have to enable it specifically through amix=inputs=79:normalize=1 for instance.

[CoRoNe]

Or introduce an option to turn it off, amix=inputs=79:normalize=0, at least.

[Elon Musk]

fixed in 57651493923ea4bc76453aa18fb0ecad7926720d

[CoRoNe]

For future reference:

​5765149 introduced a sum option which was disabled by default.
--> amix=inputs=79:sum=1.

​95b854d (a week later) silently renames this option to (my suggested) "normalize" which is enabled by default.
--> amix=inputs=79:normalize=0.